The Best Agent Prompt Is a Lint Error
Agent failures aren't intelligence problems — they're feedback problems. TypeScript strict, custom lint rules, and error messages that outperform system prompts.
AI Systems Blog
Writing on LLM agents, AI systems, behavioral engineering, and software under pressure.
-
- JSX That Outputs Markdown
Agent instructions outgrow template strings fast. jsx-md is a JSX-to-Markdown runtime — jsxImportSource, typed props, no React.
- An LLM Is Not a Deficient Mind
LLMs share the property Watts built Rorschach around — receiver-adapted output, no inner model. The engineering works when you treat that as architecture.
- Your Agent Is a Small, Low-Stakes HAL
Recurring failure modes in LLM agent systems - directive conflict, hallucination, silent fallback, and sycophancy - and the science fiction that diagnosed them first.